Predictability of the Chinese Stock Market

(1)

1

Predictability of the Chinese stock market

Author: Yuxiao Liu

Student number:

10754296

Thesis supervisor: Dr. Liang Zou

Finish date:

07-2018

UNIVERSITY VAN AMSTERDAM

AMSTERDAM BUSINESS SCHOOL

MSc Finance

(2)

2

Statement of Originality

This document is written by Yuxiao Liu who declares to take full responsibility for the

contents of this document.

I declare that the text and the work presented in this document are original and that

no sources other than those mentioned in the text and its references have been

used in creating it.

The Faculty of Economics and Business is responsible solely for the supervision of

completion of the work, not for the contents.

(3)

3

ABSTRACT

This paper examines the forecasting power of dividend yield, dividend payout ratio,

price-earnings ratio, hiring rate and spreads in the Chinese stock market. Evidences

show that dividend yield and the price-earnings ratio can help to predict short- and

long-horizon excess return, especially for B-shares; however, short-horizon

predictability occurs only after the implementation of several important financial

liberation policies, while long-horizon predictability always exists. Our research

results imply that in China's stock market, prices do not fully reflect information; this

means that even if the financial liberation policy is implemented, the market

efficiency of Chinese stocks is still insufficient. The present findings lead to a

conclusion that is contrary to recent studies about the improvement of the Chinese

market efficiency.

Keywords: Market Efficiency, the Chinese stock market

JEL Classification: G14

(4)

4

In financial economics, understanding the empirical relationship between financial markets and macroeconomic variables has always been a crucial topic. In fact, a variety of market-related variables, such as dividend yield, dividend payout, and price-earnings ratio, are reliable predictors of stock returns has been fully demonstrated in the US market (see e.g. Fama and French, 1988; Campbell, 1986; Lamont, 1998).

However, we can also see that there are few studies on the predictability of variables related to the Chinese stock market. China's stock market has unique characteristics such as market

segmentation, high speculative tendencies, overreaction and cognitive bias. These characteristics are different from those in other countries and have not been fully studied. Therefore, the Chinese stock market provides a background for various in-depth studies, as general conclusions may not hold in the Chinese market.

China's stock market is relatively new, but it is expanding at an alarming rate. According to Shanghai Stock Exchange Factbook, since the establishment of the Shanghai Stock Exchange (SSE) in 1991, the market value has grown to 28.46 trillion yuan in 2016, and the total number of listed companies has increased from 8 to 1182. Specifically, the SSE trades two types of stocks: A-shares can only be purchased and traded by Chinese investors, and are denominated in RMB, while B shares are denominated in US dollars and are limited to foreign investors before February 2001.

In a broad consensus, the government's participation in China's stock market activities has led to a lack of transparency in the financial environment; therefore, this stock market is considered to be driven primarily by market rumors and individual investor sentiment. Early research on China's stock market showed that the market's performance in the first decade violated the assumption of weak-form efficiency (see Groenewold, Tang and Wu, 2003; Lim, Habibullah and Hinich, 2009). Nevertheless, since 2000s, the Chinese government implemented a number of measures to reform the stock market. After implementing a financial liberalization policy in February 2001, both domestic and foreign investors secured the opportunity to invest in B-shares. In addition, although there are still some restrictions after the 2005’s reform policy, qualified foreign institutional investors (QFII) is eligible to trade domestic renminbi-denominated securities. Meanwhile, another reform policy announced in 2005 allowed the conversion of non-tradable shares (NTS) into non-tradable shares (TS) by paying compensation to TS holders.Recent

(6)

6

research confirms that China's regulatory and policy reforms are promoting market integration and improving information efficiency(see for example, Chiang, Nelling and Tan, 2008; Beltratti, Bortolotti) And Caccavaio, 2016). On the other hand, recent papers focus on the behavior of Chinese stock market investors as well, such as investor herd behavior in a segmented stock market.

This paper extends the literature in two ways. Firstly, it documents the fact that dividend yield and the price-earnings ratio can help to forecast short- and long-horizon excess return, especially for B-shares; however, short-horizon predictability occurs only after the implementation of several important financial liberation policies, while long-horizon predictability always exists. The second contribution of the paper is to provide some insights into information efficiency in Chinese market. Our research results display that in China's stock market, prices do not fully reflect information; this means that even if the financial liberation policy is implemented, the market efficiency of Chinese stocks is still insufficient. The present findings lead to a conclusion that is contrary to recent studies on the improvement of weak-form efficiency in the Chinese market.

The organization of this paper is as follows. The second section introduces the review of related literature. Section 3 describes the methodological issues, Section 4 summarizes the data, and Section 5 presents the regression analysis of excess returns for both A-and B-shares. Finally, Section 6 contains the robustness check and Section 7 provides the conclusion.

2. Literature review

2.1 The Efficient Market Theory

An efficient market means that in a stock market, prices fully reflect all available information. Specifically, the price can be freely changed according to the relevant information, and the information can be fully disclosed. According to this assumption, the price of the stock market is unpredictable, and any technical analysis of the stock is invalid. Fama (1970) reviewed the empirical evidence of effective market theory and classified the effective market into three forms: weak-form efficiency, indicating that historical information cannot predict future excess returns. Any technical analysis based on past price metrics will be inaccurate; semi-strong form

(7)

7

efficiency, which implies that public information is evidently incapable of forecasting future excess returns. Any fundamental analysis based on public information of financial statements and economic conditions will be inaccurate. Lastly, strong-form efficiency, which means that the stock price fully reflects all the information that has been disclosed and undisclosed. The strong-form effective market hypothesis is recommended as a benchmark against which the importance of deviating from market efficiency can be judged. The efficiency of the Chinese stock market is crucial to market participants and the entire national economy, since the effective stock market ensures the liquidity of the financial environment, which is conducive to the stability of the national economy.

2.2 Stock Return Predictors

Traditionally, dividends and returns have been used to predict stock prices, as past studies have confirmed that they capture information that exceeds the information contained in current stock price levels. Consequently, dividends and returns are considered to have the ability to predict returns. Fama and French (1988) contented that the predictive power of dividend yields in stock returns increases as the range of regression increases. The principle of the predicted return based on dividend yield is that when the discount rate and the rate of expected return are higher, the stock price is lower relative to the dividend, and vice versa; in short, the dividend yield varies with the expected return. The author interprets the positive correlation between predictive power and time horizon by positive autocorrelation of expected returns. In addition, their findings suggest that the most influential factor in the change in returns is the change in expected return on earnings. At the same time, Lamont (1998) pointed out that in the short term, the aggregate dividend payout ratio has a substantive explanatory power because both dividends and returns have separately identifiable predictive power, but only the stock price exhibits the ability to predict long-term returns. Lamont proposes two possible reasons why dividend payout rates can predict returns. Firstly, due to the management behavior in the dividend setting, dividends measure the permanent component of stock prices, while high dividends predict high future returns. Secondly, the level of earnings is a good measure of current business conditions. Owing to risk premiums, investors require higher expected returns in a recession and lower expected returns during the boom period; while high earnings predict low returns. On the other hand,

(8)

8

Campell (1986) stated that the excess return is reliably positive only at the short term structure. Moreover, the findings of Basu (1977) showed that, on average, a portfolio with a lower price-to-earnings (P/E) ratio yields a higher absolute return after risk adjustment. Because of the inverse relationship between P/E ratio and portfolio return, publicly available P/E ratio information may ensure that investors receive risk-adjusted high returns. Accordingly, the P/E ratio may be a reliable predictor to explain the abnormal returns in the stock market.

However, an increasing number of recent papers claimed that significant evidence of predictability is mainly reflected in the prediction of short-term returns, rather than the

prediction of long-term returns. For example, Ang and Bekaert (2006) and Lettau and Ludvigson (2001). They raised a question about predictability, that is, most of the research's focus is on the predictive power of dividend yields in the long-term perspective. Their findings suggest that the statistical significance of long-term predictive power depends on the choice of standard error critically and the ability to predict future returns is not robust at different times or countries. Ang and Bekaert (2006) also suggested that the most reliable predictor of future excess returns is short-term interest rates, which is essentially available in the short horizon.

In addition, the impact of labor market friction on predicting future returns has also been studied. Belo, Lin, and Bazdresch (2014) pointed out that hiring decisions are often forward-looking because the decision of dismissal and hiring employees is costly, so hiring rates may inform the company's future value. The authors' survey results displayed that a relatively high hiring rate means that the average return rate of stocks in the future is low. This predictability is interpreted as a balanced outcome that reflects the relatively low macroeconomic risks of these companies.

2.3 The Weak-form Efficiency of Chinese Stock Markets

In Groenewold, Tang and Wu’s paper (2003), the performance of A-shares and B-shares traded on the Shanghai and Shenzhen stock exchanges between 1992 and 2001 has been assessed. Experimental evidence suggested that the value of stocks can be predicted by analysis of past values, which is clearly violation of the hypothesis of weak-form efficiency. At the same time, the authors suggested that thin trading may be the cause of this apparent predictability. In addition, Kang, Liu, and Ni (2002) argued in a study of investment strategies that overreaction to

(9)

company-9

specific information leads to short-term contrarian profits, while negative cross-serial correlation leads to intermediate-term momentum profits. And the use of contrarian strategies and

strategies can yield statistically significant abnormal returns from 1993 to 2000. This finding also violates the assumptions of efficient markets and confirms the inefficiency of China's early stock markets.

Following the market's ineffective research, Lim, Habibullah, and Hinich (2009) considered the impact of thin trading and overcame the shortcomings of China's early market efficiency studies to further analyze the early market. Their findings contented that there is still evidence that, at certain times, new information is not fully reflected in the stock prices of the Shanghai and Shenzhen stock exchanges; after correcting the thin trading impact of the transactions, it was found that China's two trading markets were effective most of the time, but not all. Another interesting finding in their paper is that the continuous dependence of China's two trading markets has been closely followed since 1997. This finding suggested that fiscal policy and economic change have similar effects on both markets.

On the other hand, recent studies show that the efficiency of the Chinese market has increased after the implementation of the financial liberation policies. Chiang, Nelling and Tan (2008) analyzed the speed of price adjustment between A-shares and B-shares. The empirical results revealed that before 2001, the stock price of A-shares adjusted faster than the B-shares. However, following the A-share liberation policy in 2001, the adjustment speed of B-shares accelerated, resulting in the corresponding difference between A-shares and B-shares reduced. Another related finding is that A-shares reacts more rapidly when bad news occurs, and when the good news happens, the B-shares exhibit a faster speed of adjustment. The evidence provided by this study shows that the adoption of the A-share liberation policy has increased the market efficiency.

An event study was conducted by Beltratti, Bortolotti and Caccavaio (2016) to assess the impact of the 2005-2006 structural reform policy on stock returns. After the policy is passed, NTS can convert to TS by means of compensation payments. The authors found evidence that the reason behind the positive abnormal returns that occurred in the few days before the

announcement of the reform process, as well as in the ten days after the readmission to trading of participating companies following the determination of the compensation was the information

(10)

10

leakage rather than the compensation risk premium. They concluded that this structural reform policy has played a significant role in improving the efficiency of China's stock market.

In addition, Chinese investors’ herding behavior has been studied with the background of a market segmentation environment (Yao, Ma, and He, 2014). The results displayed that investors exhibit different levels of herding behavior, and the herding effect in the B-share market is particularly strong. However, this herding behavior lessens over time. The authors confirmed that changes in investors’ herding behavior indicate that regulatory reforms are improving information efficiency and market integration.

As for the studies of predictability in the Chinese stock market. 18 firm-specific variables that have been well studied in the U.S market were introduced by Chen, Kim, Yao, and Yu (2010). Then, they examined their relationship to Chinese stock returns. Their findings suggest that tests of US stock returns revealed more predictors that are relatively weak in predicting China's stock market. They also provide two empirically supported explanations for the weak predictability in the Chinese market: firstly, the forecasting factors in U.S. may be more evenly distributed than China's heterogeneity; secondly, U.S. stock prices reveal more effective information than Chinese stock prices.

3. Methodological issues

3.1 Predictability regressions

Economists have a long tradition of experimenting with predicting stock market returns. In a classical specification regression, the stock market yield is dependent variable, and the lagged predictor is independent variable. The main regression considered here is 𝑟𝑡+1:𝑡+𝑘= 𝛼 + 𝛽 ∗

𝑋𝑡+ 𝜖𝑡+1:𝑡+𝑘, where 𝑟𝑡+1:𝑡+𝑘 is the one period excess return, 𝑋𝑡 is the variable of interest to

test whether it has predictive ability, 𝜖𝑡+1:𝑡+𝑘 is an error term that is not addressed in this

article, and 𝛽 measured the significance of X in predicting the equity premium. In this paper, a period is either a quarter or a year. The X variables investigated in this paper are dividend yield, dividend payout ratio, price-earnings ratio, hiring rate and spreads. All returns are continuously compounded. The null hypothesis of this paper that the Chinese stock market is not predictable in the long run (𝛽 = 0) will be tested by estimating the regression with OLS method and compute

(11)

11

standard errors following Newey and West (1987). Using the formula ∑ 𝑁𝑊 =

𝑇 𝑇−𝐾(𝑋 ′_𝑋)−1_Ω(𝑋′_{𝑋) where Ω =} 𝑇 𝑇−𝐾{∑ 𝑢1 2 𝑇 𝑡=1 𝑥1𝑥1′+ ∑𝑞𝑣=1[(1 − 𝑣 𝑞+1) ∑ 𝑥𝑡𝑢𝑡𝑢𝑡−𝑣𝑥𝑡−𝑣 𝑇 𝑡=𝑣+1 ′ + 𝑥𝑡−𝑣𝑢𝑡−𝑣𝑢𝑡𝑥𝑡′]}, we can obtain the estimator to overcome the issue of the error terms often

being found to be correlated over time in time series data since the Newey-West standard error is more robust in time series.

3.2 Out-of-sample predictability

Although financial economists have demonstrated in their previous explorations that some variables can predict stock returns, the following reasons lead to how to choose the correct predictive regression specification is still an unconfirmed problem. The first is the uncertainty of the model. Many potentially predictable economic variables are vulnerable to institutional changes, policy shocks, information technology advances, and structural instability caused by investor learning. Moreover, highly complex uncertainties make it difficult for a single predictive regression model to obtain a reliable continuous evolution of expected equity returns. Given the uncertainty and instability of this environment, predictions from a single model may produce reliable predictions over a specific time period, but it is not possible to produce consistent and reasonable predictions over a long period.

Instability of the parameters is the second reason, the diversity of potential predictors also makes it difficult to get a convincing explanation. For instance, based on a specific set of explanatory variables, one may identify statistically significant economic variables, but the significance of these variables is usually not widespread. As the available time series are usually limited, the instability of the parameters is particularly relevant. In theory, an all-inclusive specification can be built on an extremely large sample, which can include all potential predictors. Therefore, when this particular predictive regression is performed, the slope

coefficient estimates of unrelated variables converge to zero (their true values).However, in fact, when studying the topic of predictability, the number of observations of the sample is limited compared to a large number of potential variables that may have explanatory power. Therefore, one drawback of the single predictive regression paradigm is that it does not help identify which predictors are truly explanatory.

(12)

12

Most existing studies that claim to find substantial evidence that returns are predictable focus on in-sample testing. Nevertheless, as emphasized by Welch and Goyal (2008), a persistent systemic problem with predictive regression is that the predictions in out-of-sample tests are not convincing. The authors claimed that a series of predictors studied in existing literature has no significant advantage over the predictive performance of stock premiums relative to simple forecasts based on historical averages, and they concluded that structural instability was responsible for the unsustainability of individual predictive regression models to provide

significant out-of-sample performance. The comprehensive findings in their paper are a summary of relatively few out-of-sample test studies for predictability of returns, and their empirical evidence confirms the poor performance of the regression predictions presented in these studies in out-of-sample tests. For instance, Bossaerts and Hillion (1999) tested the predictive power of a range of variables in industrialized countries and failed to find robust evidence of predictive power in out-of-sample tests. Subsequently, Goyal and Welch (2003) confirmed that even in the U.S. market, the dividend-price ratio did not provide reliable empirical evidence to demonstrate predictive power in out-of-sample tests.

In fact, the lack of persistent sample evidence indicates that existing forecasting methods can not ensure the reliability of forecasting, so the improvement of forecasting methods is necessary. The forecasting power of economic variables will be significantly reduced due to the inclusion of model uncertainty and prediction instability, and the criteria for any predictive model should be whether it exhibits robust predictive power in out-of-sample tests.

3.3 Forecast combinations and economically motivated restrictions

A single, relatively simple predictive regression model cannot meet the requirements of highly uncertain and new data generation environments. Therefore, using a single predictive regression model to approximate the excess returns may be inaccurate. There are several ways to help reduce the risk of uncertainty/instability due to relying on a single regression model.

The first method is forecast combination. Over-fitting may result in poor performance of stock returns forecasts in out-of-sample tests. Reducing the model uncertainty or the risk of instability associated with a single model can be achieved by combining individual predictors (Rapach,

(13)

13

Strauss and Zhou, 2010). The combination formula can be summarized as: 𝑟_𝑡+1𝑝𝑜𝑜𝑙=

∑𝑘𝑖=1𝑤𝑖,𝑡𝑟𝑖,𝑡+1, where 𝑤𝑖,𝑡 represents the ex-ante combining weights ,and r𝑡+1 are weighted

averages of the N individual forecasts. And they also concluded that simple combinatorial methods usually outperform more complex ones. In contrast to Welch and Goyal’s findings (2008) about the poor performance of individual predictive regression models, forecast combination method generated robust evidence of the out-of-sample forecasting power over certain periods.

On the one hand, the forecast variance can be significantly reduced by the combination of predictions, because the predictions of individual prediction regression models usually seem too volatile to accurately represent the potential changes. On the other hand, predictive

combinations can also include economic variables that are often overlooked, while predictions based on historical averages are often too smooth because they ignore potential information in economic variables. In general, forecast combination are balanced by integrating information from many economic variables while avoiding overly noisy predictions.

Another method is economically motivated restrictions. Out-of-sample regressions are based on fewer observations than in-sample regressions, which means parameters are less accurately estimated; this characteristic helps to impose restrictions. The use of financial theory can restrict the regression of predicted stock returns in two ways. One is to reduce the amount of parameters for free estimation, and the other is to limit the estimate to a reasonable range. Campbell and Thompson (2008) argued that in predictive regression, if the key predictors are imposed with simple constraints in accordance with the investment theory, the out-of-sample performance of prediction will be significantly improved. Moreover, they illustrated two alternative restrictions that can be applied to any predictive regression based on investment theory: firstly, that the regression coefficient has the desired sign in theory; secondly, that the equity premium has a positive fitted value. And the authors imposed alternative restrictions in their studies first sequentially and then together, empirical evidence showed that after applying the above-mentioned economic constraints, the results of predictive regression in the out-of-sample test have been significantly improved.

(14)

14

4. Data and descriptive statistics

For financial data, we consider returns on the SSE A-Share Index and SSE B-Share Index as our measure of financial asset returns on A-shares and B-shares stocks respectively. Prices, dividends per share, earnings per share and number of employees are obtained from WIND. The risk-free rate is the yield of China’s 3-month Government Bond. The sample period is 2002-2016. Due to data availability considerations, all variables are quarterly.

Excess returns are total stock returns minus the return on China’s 3-month Government Bond. Log price (p) is the natural logarithm of the SSE Index price level. Log dividends (d) are the natural logarithm of the sum of the past four quarters of dividends per share. Log earnings (e) are the natural logarithm of a single quarter’s earnings per share.

Table 1 Summary statistics, Index A 2002Q1-2016Q4

𝑅𝑚,𝑡 𝑅𝑚,𝑡− 𝑅𝑓,𝑡 ln (𝑑𝑡/𝑝𝑡) ln (𝑑𝑡/𝑒𝑡) ln (𝑝𝑡/𝑒𝑡) 𝐻𝑁𝑡

Panel A: Summary statistics for Index A

Mean 0.0107 －0.0130 －3.5078 5.7582 9.2660 0.1166

Standard deviation 0.1629 0.1645 0.4083 0.5843 0.6299 0.0898

Min －0.4158 －0.4458 －4.6090 4.8404 8.2198 0.0345

Max 0.4251 0.4052 －2.8649 7.2974 10.9871 0.3498

Autocorrelation 0.2138 0.2211 0.8901 0.0255 0.1872 0.5273 Panel B: Correlation matrix for Index A

𝑅𝑚,𝑡 1 0.9985 －0.1738 0.1228 0.2917 －0.3074 𝑅𝑚,𝑡− 𝑅𝑓,𝑡 0.9985 1 －0.1833 0.1307 0.3085 －0.2973 ln (𝑑𝑡/𝑝𝑡) －0.1738 －0.1833 1 0.5161 －0.7733 －0.0474 ln (𝑑𝑡/𝑒𝑡) 0.1228 0.1307 0.5161 1 0.1439 0.6972 ln (𝑝𝑡/𝑒𝑡) 0.2917 0.3085 －0.7733 0.1439 1 0.1275 𝐻𝑁𝑡 －0.3074 －0.2973 －0.0474 0.6972 0.1275 1

Panel A: Summary statistics for SSE Index A. 𝑅𝑚,𝑡+1 is market return. 𝑅𝑚,𝑡+1− 𝑅𝑓,𝑡+1 is quarterly log excess

returns, calculated as total returns on the SSE Index A minus total return on 3-month government bond as of the last day of quarter t. 𝑑𝑡 is the log of dividends per share paid out in the four quarters, including quarter t. 𝑒𝑡 is

(15)

15

the log of earnings per share in quarter t. 𝑝𝑡 is the log of the index price level. 𝐻𝑁𝑡 is the ratio between net

hiring and average hiring.

Panel B: correlation matrix for Index A. The sample period is from 2002 to 2016.

The log of the dividend yield ln (𝑑𝑡/𝑝𝑡) is the difference between the log of dividends and

the log of price. The log of the dividend payout ratio ln (𝑑𝑡/𝑒𝑡) is the difference between the log

of dividends and the log of earnings. The log price-earnings ratio ln (𝑝𝑡/𝑒𝑡) is the difference

between the log of price and the log of dividends (e.g. Lamont, 1998; Ang and Bekaert, 2006). Moreover, as discussed in Belo, Lin and Bazdresch (2014), hiring rate( 𝐻𝑁𝑡) is defined

as 𝐻𝑡/[0.5 ∗ (𝑁𝑡−1+ 𝑁𝑡)], where 𝑁𝑡 is the number of employees for listed companies in the

Shanghai exchange stock market, and net hiring 𝐻𝑡 is given by the change in the number of

employees from time t-1 to time t.

Table 2 Summary statistics, Index B 2002Q1-2016Q4

𝑅𝑚,𝑡 𝑅𝑚,𝑡− 𝑅𝑓,𝑡 ln (𝑑𝑡/𝑝𝑡) ln (𝑑𝑡/𝑒𝑡) ln (𝑝𝑡/𝑒𝑡) 𝐻𝑁𝑡

Panel A: Summary statistics for Index B

Mean 0.0115 －0.0122 －4.0176 3.3694 7.3885 0.038

Standard deviation 0.1885 0.1906 0.3610 0.6156 0.6953 0.0999

Min －0.4556 －0.4878 －4.7008 2.2962 6.2780 －0.1754

Max 0.5607 0.3645 －3.1569 4.6564 8.7330 0.1933

Autocorrelation 0.1293 0.1365 0.8445 0.0264 0.1930 －0.1073 Panel B: Correlation matrix for Index B

𝑅𝑚,𝑡 1 0.9982 －0.6653 －0.0762 0.5723 －0.2403 𝑅𝑚,𝑡− 𝑅𝑓,𝑡 0.9982 1 －0.6419 －0.0547 0.5675 －0.2539 ln (𝑑𝑡/𝑝𝑡) －0.6653 －0.6419 1 0.3716 －0.6496 －0.1471 ln (𝑑𝑡/𝑒𝑡) －0.0762 －0.0547 0.3716 1 0.4645 0.0325 ln (𝑝𝑡/𝑒𝑡) 0.5723 0.5675 －0.6496 0.4645 1 0.1669 𝐻𝑁𝑡 －0.2403 －0.2539 －0.1471 0.0325 0.1669 1

Panel A: Summary statistics for SSE Index B. 𝑅𝑚,𝑡+1 is market return. 𝑅𝑚,𝑡+1− 𝑅𝑓,𝑡+1 is quarterly log excess

(16)

16

last day of quarter t. 𝑑𝑡 is the log of dividends per share paid out in the four quarters, including quarter t. 𝑒𝑡 is

the log of the earnings per share in quarter t. 𝑝𝑡 is the log of the index price level. 𝐻𝑁𝑡 is the ratio between net

hiring and average hiring.

Panel B: correlation matrix for Index B. The sample period is from 2002 to 2016.

Tables 1 and 2 present summary statistics for the variables mentioned above in Index A and Index B for the series studied in this paper. The higher correlation between log dividend yield and excess return, as well as log price-earnings ratio and excess return in Index B, may be due to the relatively thin trading. The autocorrelation coefficient in both tables shows that the dividend yield is a highly persistent series. The high persistence of the dividend yield is due to the high persistence in both dividends and prices.

5. Empirical results

5.1 Quarterly Forecasting

First of all, we test the forecasting ability of market indicators to predict excess returns in the short horizon. We select the quarterly excess returns as the dependent variable, select the dividend yield, the dividend payout ratio and the price-earnings ratio as the independent variables to conduct the OLS regression test, respectively.

5.1.1 Predictability of Quarterly A-share Excess Returns

Table 3 replicates the results of OLS regressions using A-share quarterly data on excess returns, dividends, and earnings. The dependent variables are quarterly log excess returns. For each regression, we report the point estimates of the included explanatory variable, 𝑅2 and the Newey-West adjusted t-statistics. The first four rows of Table 3 show that when the dividend yield, dividend payout ratio, price-earnings ratio and spread are selected as individual predictors, the predictive power of the above forecasting factors on the excess returns of A-shares is not

(17)

17

significant. The last row of table 3 displays that the combination of payout ratio and price-earnings ratio also does not exhibit any forecasting power.

Table 3 Forecasting Quarterly A-share Excess Returns, 2002-2016

Constant ln (𝑑𝑡/𝑝𝑡) ln (𝑑𝑡/𝑒𝑡) ln (𝑝𝑡/𝑒𝑡) 𝑆𝑝𝑟𝑒𝑎𝑑𝑡 𝑅2 1 log(D/P) 0.252 0.075 0.034 (0.243) (0.072) 2 log(D/E) －0.024 －0.002 0.001 (0.167) (0.028) 3 log(P/E) 0.259 －0.029 0.012 (0.193) (0.033) 4 spread －0.016 0.304 0.001 (0.05) (3.78) 5 log(D/E) log(P/E) 0.371 0.08 －0.094 2.471 0.043 (0.35) (0.079) (0.083) (4.512)

Regression of quarterly excess stock return of A-shares on lagged dividend yields, dividend payout ratios and price-earnings ratios, 2002Q1-2016Q4. The dependent variable 𝑅𝑚,𝑡− 𝑅𝑓,𝑡 is quarterly log excess returns on the

SSE A-share Index. ln (𝑑𝑡/𝑝𝑡) is the log dividend yield, ln (𝑑𝑡/𝑒𝑡) is the log dividend payout ratio, ln (𝑝𝑡/𝑒𝑡) is

the log price-earnings ratio, and 𝑆𝑝𝑟𝑒𝑎𝑑𝑡 is the difference between the yield of China’s 10-year Government

Bond and the yield of China’s 3-month Government Bond. Numbers in parentheses show the Newey-West adjusted t-statistics. *, **, and *** indicate significance at 10%, 5%, and 1%, respectively.

In order to explore the changes in the predictability of the Chinese stock market before and after the implementation of the financial emancipation policy, we divided the time period into two parts: 2002-2007 and 2008-2016. Hence, we run the same regressions in table 3 for the two time periods respectively. In table 4, we show the regression results of forecast factors over quarterly excess returns during the period 2002-2007, while table 5 displays the results for the period 2008-2016.

From table 4, it can be seen that for the period 2002-2007, all forecasting indicators are not significant and do not predict excess returns as well.

(18)

18 Table 4 Forecasting Quarterly A-share Excess Returns, 2002-2007

Constant ln (𝑑𝑡/𝑝𝑡) ln (𝑑𝑡/𝑒𝑡) ln (𝑝𝑡/𝑒𝑡) 𝑆𝑝𝑟𝑒𝑎𝑑𝑡 𝑅2 1 log(D/P) －0.217 －0.07 0.026 (0.31) (0.086) 2 log(D/E) 0.184 －0.026 0.008 (0.29) (0.049) 3 log(P/E) －0.005 0.004 0.001 (0.376) (0.04) 4 spread 0.11 －5.321 0.018 (0.135) (9.041) 5 log(D/E) 0.162 －0.093 0.054 －6.102 0.053 log(P/E) (0.66) (0.117) (0.101) (11.09)

However, when period 2002-2007 is excluded, as table 5 establishes, dividend yield and price-earnings ratio show significant predictive power for short-term excess returns. The first four rows show that dividend yield is indeed a better forecaster than dividend payout ratio, price-earnings ratio and spread, since the dividend yield has more explanatory power than other indicators.

When price-earnings ratio and dividend payout ratio are used as separate predictors, although both price-earnings ratio and dividend payout ratios are negatively correlated with future returns, only P/E ratios show significant predictive power. Row 5 shows that when both dividend yield and price-earnings ratio are included, the results appear unexpected. Conditional on the price-earnings ratio, the dividend yield is positively correlated with future returns; moreover, the forecasting power for both indicators is more statistically significant. The regression results show

(19)

19

that the unpredictability of the A-share in the full time period is mainly affected by the period from 2002-2007. After the comprehensive implementation of the financial emancipation policy, the predictability of A-shares improves significantly after 2008.

Table 5 Forecasting Quarterly A-share Excess Returns, 2008-2016

Constant ln (𝑑𝑡/𝑝𝑡) ln (𝑑𝑡/𝑒𝑡) ln (𝑝𝑡/𝑒𝑡) 𝑆𝑝𝑟𝑒𝑎𝑑𝑡 𝑅2 1 log(D/P) 0.508 0.157** 0.165 (0.249) (0.074) 2 log(D/E) －0.022 －0.003 0.001 (0.184) (0.03) 3 log(P/E) 0.75 －0.086* 0.094 (0.4) (0.045) 4 spread －0.028 －1.159 0.002 (0.056) (4.963) 5 log(D/E) 1.276 0.189*** －0.27*** 7.925 0.27 log(P/E) (0.386) (0.064) (0.078) (6.485)

5.1.2 Predictability of Quarterly B-share Excess Returns

To explore the potential differences between A-shares and B-shares as regards predictability, we conduct the same analysis of B-shares as for A-shares.

(20)

20

Table 6 replicates the results of OLS regressions using B-share quarterly data for the full period from 2002-2016, while table 7 displays the OLS regression results for the period 2002-2008. The estimates of the included explanatory variable, 𝑅2_{and the Newey-West corrected t-statistics}

have been displayed. We can find that the explanatory power of all forecasting indicators is weak and the predictability is not significant for either the full period or the period from 2002-2007. This result is consistent with the results for A-shares.

Table 6 Forecasting Quarterly B-share Excess Returns, 2002-2016

Constant ln (𝑑𝑡/𝑝𝑡) ln (𝑑𝑡/𝑒𝑡) ln (𝑝𝑡/𝑒𝑡) 𝑆𝑝𝑟𝑒𝑎𝑑𝑡 𝑅2 1 log(D/P) 0.475 0.121 0.052 (0.323) (0.08) 2 log(D/E) －0.005 －0.002 0.001 (0.068) (0.021) 3 log(P/E) 0.028 －0.006 0.002 (0.084) (0.012) 4 spread －0.028 1.470 0.002 (0.05) (3.961) 5 log(D/E) 0.176 0.054 －0.051 0.474 0.028 log(P/E) (0.132) (0.068) (0.042) (4.269)

Regression of quarterly excess stock return of B-shares on lagged dividend yields, dividend payout ratios and price-earnings ratios, 2002Q1-2016Q4. The dependent variable 𝑅𝑚,𝑡− 𝑅𝑓,𝑡 is quarterly log excess returns on the

SSE B-share Index. ln (𝑑𝑡/𝑝𝑡) is the log dividend yield, ln (𝑑𝑡/𝑒𝑡) is the log dividend payout ratio, ln (𝑝𝑡/𝑒𝑡) is

(21)

21 Table 7 Forecasting Quarterly B-share Excess Returns, 2002-2007

Constant ln (𝑑𝑡/𝑝𝑡) ln (𝑑𝑡/𝑒𝑡) ln (𝑝𝑡/𝑒𝑡) 𝑆𝑝𝑟𝑒𝑎𝑑𝑡 𝑅2 1 log(D/P) －0.11 －0.033 0.006 (0.364) (0.098) 2 log(D/E) 0.016 －0.002 0.001 (0.084) (0.027) 3 log(P/E) －0.02 0.004 0.003 (0.081) (0.014) 4 spread 0.133 －7.762 0.028 (0.148) (9.85) 5 log(D/E) 0.187 －0.118 0.05 －7.713 0.076 log(P/E) (0.176) (0.113) (0.056) (9.412)

Turning to the period 2007-2016, as table 8 establishes, dividend yield and price-earnings ratio also show significant predictive power for short-term excess returns in B-shares and confirms that the dividend yield is also a better forecaster in B-shares than other indicators; moreover, the dividend yield shows more significant predictive power in B-share returns than in A-share returns. As Row 3 indicates, the price-earnings ratio is also significant; however, contrary to the performance of this indicator in A-shares, the price-earnings ratio is positively correlated with the future returns of B-shares.

Row 5 shows that when the dividend yield and price-earnings ratio are combined, the dividend yield is positively correlated with future returns while the price-earnings ratio is negatively correlated with future returns and both indicators display significant forecasting power. In

(22)

22

addition, a comparison of regression results indicates that the unpredictability of stock returns throughout the period was mainly affected by the period 2002-2007.Forecasting indicators have more explanatory power in the market of B-shares.

Table 8 Forecasting Quarterly B-share Excess Returns, 2007-2016

Constant ln (𝑑𝑡/𝑝𝑡) ln (𝑑𝑡/𝑒𝑡) ln (𝑝𝑡/𝑒𝑡) 𝑆𝑝𝑟𝑒𝑎𝑑𝑡 𝑅2 1 log(D/P) 1.691 0.415*** 0.268 (0.319) (0.078) 2 log(D/E) －0.01 －0.005 0.001 (0.119) (0.035) 3 log(P/E) 0.402 0.058* 0.041 (0.225) (0.03) 4 spread －0.048 1.982 0.005 (0.054) (4.71) 5 log(D/E) 1.686 0.408*** －0.41*** 0.815 0.269 log(P/E) (0.349) (0.097) (0.081) (3.797)

5.1.3 Quarterly Forecasting power of Prices, Dividends, and Earnings

To test the separately identifiable forecasting ability of prices, dividends and earnings in the short horizon, we regress the quarterly excess stock return of both A-shares and B-shares on lagged prices, lagged dividends and lagged earnings.

(23)

23

In Table 9, Panel A shows the regression results of A-shares, while panel B displays the results of B-shares. As can be seen from Rows 1 and 4, only prices have significant predictive power, and the relationship between price and future excess returns is negatively correlated. One possible explanation for this relationship is the discount rate, when the required future rate of return is high, the stock price is low, this means that today's low prices predict high returns in the future.

Table 9 Quarterly Excess Return Forecasting Using Prices, Dividends, and Earnings

Constant 𝑝𝑡 𝑑𝑡 𝑒𝑡 𝑅2

Panel A: Quarterly forecasting for A-Share

1 0.098 －0.00004* 0.071 (0.061) (0.00003) 2 －0.004 －0.0001 0.001 (0.055) (0.0005) 3 －0.008 －0.015 0.001 (0.045) (0.127)

Panel B: Quarterly forecasting for B-Share

4 0.074 －0.0004* 0.042 (0.061) (0.0002) 5 0.002 －0.003 0.001 (0.097) (0.023) 6 －0.011 0.009 0.001 (0.044) (0.179)

Regression of quarterly excess stock return of both A-shares and B-shares on lagged prices, lagged dividends and lagged earnings, 2002Q1-2016Q4. The dependent variable 𝑅𝑚,𝑡− 𝑅𝑓,𝑡 is quarterly log excess returns, 𝑝𝑡 is the

log of index price level, 𝑑𝑡 is the log dividend, and 𝑒𝑡 is the log earnings. Numbers in parentheses show the

Newey-West adjusted t-statistics. *, **, and *** indicate significance at 10%, 5%, and 1%, respectively.

(24)

24

5.2 Annual Forecasting

5.2.1 Predictability of Annual Excess Returns

Next, to test the predictive power of the indicators in this paper for long-term returns, we conducted the same regression test for long-term returns as what we did for short-term returns.

Table 10 Forecasting Annual A-share Excess Returns, 2002-2016

Constant ln (𝑑𝑡/𝑝𝑡) ln (𝑑𝑡/𝑒𝑡) ln (𝑝𝑡/𝑒𝑡) 𝑆𝑝𝑟𝑒𝑎𝑑𝑡 HN𝑡 𝑅2 1 －0.97 －0.261** 0.069 (0.4) (0.128) 2 －0.33 0.048 0.004 (0.372) (0.067) 3 －1.612 0.169* 0.06 (0.833) (0.095) 4 －0.176 10.16 0.027 (0.148) (10.77) 5 0.073 －0.904 0.029 (0.147) (0.622) 6 －1.697 －0.178 0.285* 3.126 0.419 (0.773) (0.153) (0.167) (11.09)

Regression of Annual excess stock return of A-shares on lagged dividend yields, dividend payout ratios and price-earnings ratios, 2002Q1-2016Q4. The dependent variable 𝑅𝑚,𝑡− 𝑅𝑓,𝑡 is the cumulative log excess returns over

the last four quarters on the SSE A-share Index. ln (𝑑𝑡/𝑝𝑡) is the log dividend yield , ln (𝑑𝑡/𝑒𝑡) is the log

dividend payout ratio, ln (𝑝𝑡/𝑒𝑡) is the log price-earnings ratio, and 𝑆𝑝𝑟𝑒𝑎𝑑𝑡 is the difference between the yield

of China’s 10-year Government Bond and the yield of China’s 3-month Government Bond, 𝐻𝑁𝑡 is the ratio

between net hiring and average hiring. Numbers in parentheses show the Newey-West adjusted t-statistics. *, **, and *** indicate significance at 10%, 5%, and 1%, respectively.

(25)

25

Row 1 of tables 10 and 11 shows that in the forecast of long-term returns, the dividend yield in the A-shares and B-shares also shows a stronger explanatory power than other indicators, and the dividend yield is more significant for the B-shares than for the A-shares. However, in the prediction of long-term excess return, the dividend yield is negatively correlated with the future return, which is contrary to the forecast of dividend yield on short-term return. Row 3 of tables 10 and 11 shows that the price-earnings ratio is only significant for A-shares in the long horizon. As shown by Rows 2, 4 and 5 of tables 10 and 11, dividend ratio, spread and hiring rate do not exhibit significant forecasting power for both A and B-shares in the long horizon.

Row 6 of tables 10 and 11 indicates that, when the dividend yield and the price-earnings ratio are combined to predict the long-term excess return, there is a positive correlation between the dividend yield and the future rate of return, and the price-earnings ratio is negatively

correlated with the future rate of return. Although the forecasting power for both indicators is statistically significant, the predictive ability of this combination in B-shares has more explanatory power than is the case for A-shares.

Table 11 Forecasting Annual B-share Excess Returns, 2002-2016

Constant ln (𝑑𝑡/𝑝𝑡) ln (𝑑𝑡/𝑒𝑡) ln (𝑝𝑡/𝑒𝑡) 𝑆𝑝𝑟𝑒𝑎𝑑𝑡 HN𝑡 𝑅2 1 －1.558 －0.38*** 0.094 (0.444) (0.106) 2 －0.158 0.037 0.004 (0.161) (0.054) 3 －0.49* 0.063 0.027 (0.293) (0.045) 4 －0.175 11.39 0.027 (0.147) (12.97) 5 0.004 －0.788 0.017 (0.171) (1.038) 6 －0.92* －0.316*** 0.237** 17.14 0.565 (0.501) (0.107) (0.09) (11.5)

(26)

26

Regression of Annual excess stock return of B-shares on lagged dividend yields, dividend payout ratios and price-earnings ratios, 2002Q1-2016Q4. The dependent variable 𝑅𝑚,𝑡− 𝑅𝑓,𝑡 is the cumulative log excess returns over

the last four quarters on the SSE B-share Index. ln (𝑑𝑡/𝑝𝑡) is the log dividend yield , ln (𝑑𝑡/𝑒𝑡) is the log

dividend payout ratio, ln (𝑝𝑡/𝑒𝑡) is the log price-earnings ratio, 𝑆𝑝𝑟𝑒𝑎𝑑𝑡 is the difference between the yield of

China’s 10-year Government Bond and the yield of China’s 3-month Government Bond, and 𝐻𝑁𝑡 is the ratio

between net hiring and average hiring. Numbers in parentheses show the Newey-West adjusted t-statistics. *, **, and *** indicate significance at 10%, 5%, and 1%, respectively.

5.2.2 Annual Forecasting power of Prices, Dividends, and Earnings

Table 12 Annual Excess Return Forecasting Using Prices, Dividends, and Earnings

Panel A: Quarterly forecasting for A-Share

1 －0.35 0.0001*** 0.085 (0.101) (0.00003) 2 －0.015 －0.0005 0.002 (0.175) (0.002) 3 0.004 －0.208 0.006 (0.149) (0.347)

Panel B: Quarterly forecasting for B-Share

4 －0.284 0.001* 0.065 (0.187) (0.0006) 5 －0.161 0.036 0.005 (0.373) (0.086) 6 －0.016 －0.148 0.001 (0.164) (0.648)

Regression of annual excess stock return of both A-shares and B-shares on lagged prices, lagged dividends and lagged earnings, 2002Q1-2016Q4. The dependent variable 𝑅𝑚,𝑡− 𝑅𝑓,𝑡 is the cumulative log excess returns over

(27)

27

Numbers in parentheses show the Newey-West adjusted t-statistics. *, **, and *** indicate significance at 10%, 5%, and 1%, respectively.

We also regress annual excess stock return of both A-shares and B-shares on lagged prices, lagged dividends and lagged earnings to test whether prices, dividends and returns have separate identifiable predictability for long-term excess return. In Table 12, panel A shows the regression results of A-shares, while panel B displays the results for B-shares. Rows 1 and 4 reveal that only price has significant forecasting power. However, in the long horizon, the relationship between price and future excess returns is positive. The influencing factors behind this relationship may be related to herding behavior, overreaction and cognitive bias in the stock market.

6. Robustness checks

Because the above tests are all estimated using the full sample, one of the possible problems with the predicted results is the in-sample test. Accordingly, we conduct robustness checks on the SSE Commercial Index and CSI 300 Index.

6.1 SSE Commercial Index

The sample stock of the Shanghai Stock Exchange Commercial Index is all listed stocks in the commercial industry (including both A-shares and B-shares), reflecting the commercial industry's economic conditions and the overall changes in its share price.

Table 13 Forecasting Quarterly SSE Commercial Index Excess Returns 2007-2016

Constant ln (𝑑𝑡/𝑝𝑡) ln (𝑑𝑡/𝑒𝑡) ln (𝑝𝑡/𝑒𝑡) 𝑆𝑝𝑟𝑒𝑎𝑑𝑡 𝑅2

1 log(D/P) 0.353 0.062 0.031

(0.305) (0.051)

2 log(D/E) 0.028 －0.016 0.004

(28)

28 3 log(P/E) 0.752 －0.081** 0.059 (0.336) (0.035) 4 spread －0.02 －0.762 0.001 (0.067) (5.947) 5 log(D/E) 1.326 0.084 －0.177 5.647 0.104 log(P/E) (0.715) (0.069) (0.105) (8.697)

Regression of quarterly excess stock return of SSE Commercial Index on lagged dividend yields, dividend payout ratios and price-earnings ratios, 2007Q1-2016Q4. The dependent variable 𝑅𝑚,𝑡− 𝑅𝑓,𝑡 is quarterly log excess

returns on the SSE Commercial Index. ln (𝑑𝑡/𝑝𝑡) is the log dividend yield, ln (𝑑𝑡/𝑒𝑡) is the log dividend payout

ratio, ln (𝑝𝑡/𝑒𝑡) is the log price-earnings ratio, and 𝑆𝑝𝑟𝑒𝑎𝑑𝑡 is the difference between the yield of China’s

10-year Government Bond and the yield of China’s 3-month Government Bond. Numbers in parentheses show the Newey-West adjusted t-statistics. *, **, and *** indicate significance at 10%, 5%, and 1%, respectively.

Row 1 of tables 13 and 14 show that dividend yield only displays significant predictive ability of the SSE commercial Index in the long horizon. On the other hand, Row 3 of tables 12 and 13 indicate that the price-earnings ratio only shows significant forecasting power in the short horizon. However, Row 5 of tables 13 and 14 reveal that the combination of dividend yield and price-earnings ratio does not exhibit forecasting power in the commercial industry. Overall, although the significance of the forecast indicators in the commercial market is not as

pronounced as for the overall market, these forecast indicators still exhibit a certain degree of predictability for the commercial market.

Table 14 Forecasting Annual SSE Commercial Index Excess Returns 2002-2016

Constant ln (𝑑𝑡/𝑝𝑡) ln (𝑑𝑡/𝑒𝑡) ln (𝑝𝑡/𝑒𝑡) 𝑆𝑝𝑟𝑒𝑎𝑑𝑡 𝑅2 1 log(D/P) －1.193 －0.192*** 0.053 (0.336) (0.057) 2 log(D/E) 0.221 －0.068 0.013 (0.221) (0.047) 3 log(P/E) －0.333 0.032 0.003

(29)

29 (0.692) (0.072) 4 spread －0.204 14.83 0.048 (0.187) (13.01) 5 log(D/E) －0.918 －0.165** 0.138 12.16 0.083 log(P/E) (0.602) (0.079) (0.088) (13.01)

Regression of Annual excess stock return of SSE commercial Index on lagged dividend yields, dividend payout ratios and price-earnings ratios, 2002Q1-2016Q4. The dependent variable 𝑅𝑚,𝑡− 𝑅𝑓,𝑡 is the cumulative log

excess returns over the last four quarters on the SSE Commercial Index. ln (𝑑𝑡/𝑝𝑡) is the log dividend yield,

ln (𝑑𝑡/𝑒𝑡) is the log dividend payout ratio, ln (𝑝𝑡/𝑒𝑡) is the log price-earnings ratio, and 𝑆𝑝𝑟𝑒𝑎𝑑𝑡 is the

difference between the yield of China’s 10-year Government Bond and the yield of China’s 3-month Government Bond. Numbers in parentheses show the Newey-West adjusted t-statistics. *, **, and *** indicate significance at 10%, 5%, and 1%, respectively.

We also regress excess stock return of the SSE Commercial Index on lagged prices, lagged dividends and lagged earnings to test the separately identifiable forecasting ability of prices, dividends and earnings. The regression results of the short horizon has been shown in panel A, while panel B displays the regression results of the long horizon. As can be seen from Rows 1 and 4 of table 15, price has significant forecasting power, and price is negatively correlated with future excess return in the short run but positively correlated with the future excess return in the long run. This result is consistent with the market as a whole.

Table 15 SSE Commercial Index Forecasting Using Prices, Dividends, and Earnings

Panel A: Quarterly forecasting

1 0.07 －0.00003* 0.045

(0.053) (0.00002)

2 0.004 －0.002 0.002

(0.038) (0.002)

(30)

30

(0.041) (0.172)

Panel B: Annual forecasting

4 －0.226 0.00008* 0.053 (0.175) (0.00005) 5 0.028 －0.008 0.009 (0.141) (0.009) 6 －0.048 0.126 0.001 (0.155) (0.540)

Panel A: Regression of quarterly excess stock return of SSE Commercial Index on lagged prices, lagged dividends and lagged earnings, 2002Q1-2016Q4. The dependent variable 𝑅𝑚,𝑡− 𝑅𝑓,𝑡 is quarterly log excess returns.

Panel B: Regression of annual excess stock return of SSE commercial Index on lagged prices, lagged dividends and lagged earnings, 2002Q1-2016Q4. The dependent variable 𝑅𝑚,𝑡− 𝑅𝑓,𝑡 is the cumulative log excess returns over

the last four quarters. 𝑝𝑡 is the log of index price level, 𝑑𝑡 is the log dividend, and 𝑒𝑡 is the log earnings. *, **,

and *** indicate significance at 10%, 5%, and 1%, respectively. Numbers in parentheses show the Newey-West adjusted t-statistics.

6.2 CSI 300 Index

The CSI 300 Index covers a sample of 300 A-shares on the Shanghai and Shenzhen stock exchanges, including about 60% of the market capitalization, reflecting the operating status of the Chinese securities market.

Table 16 Forecasting Quarterly CSI 300 Index Excess Returns 2007-2016

Constant ln (𝑑𝑡/𝑝𝑡) ln (𝑑𝑡/𝑒𝑡) ln (𝑝𝑡/𝑒𝑡) 𝑆𝑝𝑟𝑒𝑎𝑑𝑡 𝑅2 1 log(D/P) 0.653 0.167** 0.183 (0.263) (0.066) 2 log(D/E) －0.043 0.0008 0.001 (0.193) (0.036) 3 log(P/E) 0.796 －0.092* 0.1

(31)

31 (0.434) (0.049) 4 spread －0.033 －0.534 0.001 (0.058) (5.178) 5 log(D/E) 1.416 0.191 －0.274** 8.372 0.292 log(P/E) (0.37) (0.045) (0.061) (6.188)

Regression of quarterly excess stock return of CSI 300 Index on lagged dividend yields, dividend payout ratios and price-earnings ratios, 2007Q1-2016Q4. The dependent variable 𝑅𝑚,𝑡− 𝑅𝑓,𝑡 is quarterly log excess returns on the

CSI 300 Index. ln (𝑑𝑡/𝑝𝑡) is the log dividend yield, ln (𝑑𝑡/𝑒𝑡) is the log dividend payout ratio, ln (𝑝𝑡/𝑒𝑡) is the

log price-earnings ratio, and 𝑆𝑝𝑟𝑒𝑎𝑑𝑡 is the difference between the yield of China’s 10-year Government Bond

and the yield of China’s 3-month Government Bond. Numbers in parentheses show the Newey-West adjusted t-statistics. *, **, and *** indicate significance at 10%, 5%, and 1%, respectively.

Table 17 Forecasting Annual CSI 300 Index Excess Returns 2002-2016

Constant ln (𝑑𝑡/𝑝𝑡) ln (𝑑𝑡/𝑒𝑡) ln (𝑝𝑡/𝑒𝑡) 𝑆𝑝𝑟𝑒𝑎𝑑𝑡 𝑅2 1 log(D/P) －1.417 －0.339* 0.108 (0.672) (0.175) 2 log(D/E) 0.047 －0.007 0.000 (0.444) (0.08) 3 log(P/E) －1.67 0.186** 0.057 (0.723) (0.085) 4 spread －0.226 21.21* 0.088 (0.146) (11.62) 5 log(D/E) －1.707 －0.216 0.291 11.37 0.139 Log(P/E) (0.856) (0.217) (0.216) (12.89)

Regression of annual excess stock return of CSI 300 Index on lagged dividend yields, dividend payout ratios and price-earnings ratios, 2002Q1-2016Q4. The dependent variable 𝑅𝑚,𝑡− 𝑅𝑓,𝑡 is the cumulative log excess returns

over the last four quarters on the CSI 300 Index. ln (𝑑𝑡/𝑝𝑡) is the log dividend yield, ln (𝑑𝑡/𝑒𝑡) is the log

dividend payout ratio, ln (𝑝𝑡/𝑒𝑡) is the log price-earnings ratio, and 𝑆𝑝𝑟𝑒𝑎𝑑𝑡 is the difference between the yield

(32)

32

parentheses show the Newey-West adjusted t-statistics. *, **, and *** indicate significance at 10%, 5%, and 1%, respectively.

Row 1 of tables 16 and 17 show that dividend yield displays significant predictive ability of in the CSI 300 Index is in both the short and long horizon. Moreover, row 3 of tables 16 and 17 indicate that the price-earnings ratio shows significant forecasting power in both the short and long horizon. However, Row 5 of tables 16 and 17 reveal that the combination of dividend yield and price-earnings ratio do not exhibit forecasting power. In addition, Row 4 of table 17 shows that the predictability of spread is significant. Overall, the significance of the forecast indicators in the CSI 300 Index is similar to that for the SSE Index, as the forecast indicators show the certain forecasting power of the excess return.

Table 18 CSI 300 Index Forecasting Using Prices, Dividends, and Earnings

Panel A: Quarterly forecasting

1 0.196 －0.00007* 0.157 (0.07) (0.00002) 2 0.075 －0.002 0.021 (0.082) (0.002) 3 －0.009 0.009 0.009 (0.041) (0.172)

Panel B: Annual forecasting

4 －0.25 0.00009 0.036 (0.311) (0.0008) 5 0.243 －0.005 0.046 (0.286) (0.005) 6 0.19 －0.508 0.032 (0.229) (0.429)

(33)

33

Panel A: Regression of quarterly excess stock return of CSI 300 Index on lagged prices, lagged dividends and lagged earnings, 2002Q1-2016Q4. The dependent variable 𝑅𝑚,𝑡− 𝑅𝑓,𝑡 is quarterly log excess returns.

Panel B: Regression of annual excess stock return of CSI 300 Index on lagged prices, lagged dividends and lagged earnings, 2002Q1-2016Q4. The dependent variable 𝑅𝑚,𝑡− 𝑅𝑓,𝑡 is the cumulative log excess returns over the last

four quarters. 𝑝𝑡 is the log of index price level, 𝑑𝑡 is the log dividend, and 𝑒𝑡 is the log earnings. Numbers in

parentheses show the Newey-West adjusted t-statistics. *, **, and *** indicate significance at 10%, 5%, and 1%, respectively.

We also regress the excess stock return of the CSI 300 Index on lagged prices, lagged dividends and lagged earnings to test the separately identifiable forecasting ability. In table 18, panel A shows the regression results of the short horizon, while panel B displays the regression results of the long horizon. Rows 1 and 4 of table 18 show that price only has significant forecasting power in the short run for the CSI 300 Index.

7. Conclusion

In this article, we explore whether the Chinese stock market is predictable. Empirical evidence shows that both dividend yield and price-earnings ratio show significant explanatory power when forecasting short-term and long-term excess return. This forecasting power is mainly due to the fact that, while the price has identifiable forecasting ability, forecasting indicators have more significant forecasting ability than price alone; the predictive power of forecasting indicators for the B-shares is particularly significant. More specifically, for the prediction of short-term excess return, the significance of market predictability comes after several major financial liberation policies. The gradual evolution may be because, after these policies were implemented,

dividends and earnings contain less noise information, and can thus help to measure the value of future dividends and correlate with business conditions. For the prediction of long-term excess return, both dividend yield and price-earnings ratio display consistent forecasting power.

However, the dividend payout ratio, spread and hiring rate do not show any predictability for the Chinese market.

Our research can also provide some insights into China's stock market efficiency. Because the forecasting indicators are particularly significant in B-shares, this also confirms the findings of

(34)

34

previous studies that suggest that there are more obvious herding behaviors and cognitive biases in B-shares. Contrary to previous studies, our research shows that in China's stock market, prices do not fully reflect information. This means that even if a financial liberation policy is

implemented, the market efficiency of Chinese stocks is still not sufficient. Therefore, we questioned the improvement of Chinese market efficiency the suggested by previous research. For policymakers, the existing financial liberation policy can be seen to have had a certain effect on the improvement of market efficiency; however, the policy reform and more in-depth promotion also cannot be ignored.

(35)

35

References

Ang, A., & Bekaert, G. (2006). Stock Return Predictability: Is it There?. Review of Financial

Studies, 20(3), 651-707.

Avramov, D. (2002). Stock return predictability and model uncertainty. Journal of Financial

Economics, 64(3), 423-458.

Basu, S. (1977). Investment Performance of Common Stocks in Relation to Their Price-Earnings Ratios: A Test of the Efficient Market Hypothesis. The Journal of Finance, 32(3), 663.

Belo, F., Lin, X., & Bazdresch, S. (2014). Labor Hiring, Investment, and Stock Return Predictability in the Cross Section. Journal of Political Economy, 122(1), 129-177.

Beltratti, A., Bortolotti, B., & Caccavaio, M. (2016). Stock market efficiency in China: Evidence from the split-share reform. The Quarterly Review of Economics and Finance, 60, 125-137. Bossaerts, P., & Hillion, P. (1999). Implementing Statistical Criteria to Select Return Forecasting

Models: What Do We Learn?. Review of Financial Studies, 12(2), 405-428.

Campbell, J. (1987). Stock returns and the term structure. Journal of Financial Economics, 18(2), 373-399.

Campbell, J., & Thompson, S. (2007). Predicting Excess Stock Returns Out of Sample: Can Anything Beat the Historical Average?. Review of Financial Studies, 21(4), 1509-1531.

Chen, X., Kim, K., Yao, T., & Yu, T. (2010). On the predictability of Chinese stock returns.

Pacific-Basin Finance Journal, 18(4), 403-425.

Chiang, T., Nelling, E., & Tan, L. (2008). The speed of adjustment to information: Evidence from the Chinese stock market. International Review of Economics & Finance, 17(2), 216-229.

Fama, E., & French, K. (1988). Dividend yields and expected stock returns. Journal of Financial

Economics, 22(1), 3-25.

Groenewold, N., Tang, S., & Wu, Y. (2003). The efficiency of the Chinese stock market and the role of the banks. Journal of Asian Economics, 14(4), 593-609.

Goyal, A., & Welch, I. (2003). Predicting the Equity Premium with Dividend Ratios. Management

Science, 49(5), 639-654.

Goyal, A., & Welch, I. (2007). A Comprehensive Look at the Empirical Performance of Equity Premium Prediction. Review of Financial Studies, 21(4), 1455-1508.

(36)

36

1993–2000. Pacific-Basin Finance Journal, 10(3), 243-265.

Lamont, O. (1998). Earnings and Expected Returns. The Journal of Finance, 53(5), 1563-1587. Lettau, M., & Ludvigson, S. (2001). Consumption, Aggregate Wealth, and Expected Stock

Returns. The Journal of Finance, 56(3), 815-849.

Lim, K., Habibullah, M., & Hinich, M. (2009). The Weak-form Efficiency of Chinese Stock Markets. Journal of Emerging Market Finance, 8(2), 133-163.

Newey, W., & West, K. (1987). A Simple, Positive Semi-Definite, Heteroskedasticity and Autocorrelation Consistent Covariance Matrix. Econometrica, 55(3), 703.

Ni, Z., Wang, D., & Xue, W. (2015). Investor sentiment and its nonlinear effect on stock returns— new evidence from the Chinese stock market based on panel quantile regression model. Economic Modelling, 50, 266-274.

Rapach, D., Strauss, J., & Zhou, G. (2009). Out-of-Sample Equity Premium Prediction: Combination Forecasts and Links to the Real Economy. Review of Financial Studies, 23(2), 821-862.

Santos, T., & Veronesi, P. (2005). Labor Income and Predictable Stock Returns. Review of Financial

Studies, 19(1), 1-44.

Yao, J., Ma, C., & He, W. (2014). Investor herding behaviour of Chinese stock market. International

Predictability of the Chinese Stock Market